Building a Trading Data Center from Scratch: Lessons Learned

When we broke ground on the first NewMachine facility in Secaucus in 2018, we thought we knew what we were doing. Viktor had designed infrastructure for some of the most demanding firms on Wall Street. Our architects had built data centers for major cloud providers. And yet, the trading-specific requirements created challenges that no amount of general data-center experience fully prepared us for.

Lesson one: power density matters more than total power. Trading servers — especially those with FPGA cards — draw significantly more power per rack unit than typical enterprise workloads. Our first facility was designed for 8 kW per rack. Within a year, clients were requesting 20 kW. We retrofitted, but the lesson was clear: design for the most demanding tenant you can imagine, then double it.

Lesson two: vibration is the enemy of precision timing. PTP clock distribution at sub-100-nanosecond accuracy requires extremely stable oscillators, and those oscillators are sensitive to mechanical vibration. In our Chicago facility, we discovered that the HVAC system's compressor introduced micro-vibrations that degraded clock accuracy by 40 nanoseconds. We solved it by mounting critical timing equipment on vibration-dampening platforms and relocating the compressor to a separate mechanical room. It sounds esoteric, but for firms that care about nanosecond-precision timestamps, it was the difference between compliance and violation.

Lesson three: the NOC is the product. Clients do not interact with our switch fabric or our power distribution units — they interact with the people who answer the phone at 3 AM. Our NOC engineers are trained on trading workflows, market hours, and regulatory deadlines. They know that a network blip at 9:29 AM Eastern is not the same as one at midnight, and they triage accordingly.